First Author: I.Ramos BRAZIL
Co Author(s): B. Lopes A. Luz F. Faria Correia J. Lyra A. Machado R. Ambr
Purpose:
To test different approaches to combine corneal tomography and biomechanical parameters to distinguish normal corneas from keratoconus and from forme fruste keratoconus.
Setting:
Instituto de Olhos R. Ambrósio; Rio de Janeiro Corneal Tomography and Biomechanics and Brazilian Artifical Intelligence (BRAIN) Study Groups.
Methods:
One eye randomly selected from 119 normal patients (Group N), one eye randomly selected from 59 cases of bilateral clinical keratoconus (Group KC) and 15 eyes with normal front surface topography from cases with diagnosis of keratoconus in the contra-lateral eye (Group FFKC) were included. The pressure and waveform-derived parameters from the Ocular Response Analyzer (ORA, Reichert, Buffalo, USA), the deformation parameters from the Corvis ST (Oculus, Wetzlar, Germany) and the curvature, elevation and pachymetric tomographic parameters from Pentacam HR (Oculus, Wetzlar, Germany) were tabulated. A combined parameter was obtained for each instrument using Fisher linear discriminant analysis, considering the most accurate parameters from each instrument to detect keratoconus. Combinations of the parameters from the three instruments were performed to optimize accuracy for detection of keratoconus. These parameters were also tested to detect form fruste keratoconus.
Results:
Considering the detection of keratoconus among normals (N and KC groups), the areas under the Receiver Operating Characteristic (AUROC) curves were 0.936, 0.977 and 1.000 for the Fisher“s linear discriminant parameters created based on the analysis of ORA, Corvis and Pentacam data respectively. The AUROC, considering the N and FFKC groups were 0.909, 0.975 and 0.946 for the Fisher“s linear discriminant parameters created based on the analysis of ORA, Corvis and Pentacam data respectively. To distinguish N and KC, all combinations of Pentacam parameters with ORA and Corvis provided a virtually perfect separation with sensitivity and specificity of 100%. The AUROC, considering the N and FFKC groups were 0.985, 0.932 and 0.972 for the Fisher“s linear discriminant parameters created based on the combinations of Pentacam with Corvis, Pentacam with ORA, and Corvis with ORA respectively. The combination of data from all instruments provided AUROC of 0.973. The best combination to distinguish N and FFKC was obtained considering Corvis and Pentacam data, providing AUROC of 0.985, with sensitivity of 93.33% and specificity of 92.44%.
Conclusions:
Corneal tomography is very effective for the detection of keratoconus with a virtually perfect separation. Biomechanical data enhances the detection of mild forms of ectasia such as the cases considered as FFKC. The combinations of corneal imaging parameters that assess tomographic and biomechanical properties enhances the ability for detecting mild forms of ectasia. FINANCIAL DISCLOSURE?: ... gains financially from product or procedure presented, ... receives nonNomonetary benefits from a company producing, developing or supplying the product or procedure presented.
Please wait while information is loading.
